5.1. Trafo

Program trafo allows you to transform spectra and response matrices from standard OGIP format (or from SPEX version 1 format) to the format used by SPEX. It also gives you some options to combine data sets for simultaneous spectral fitting, and to optimises the spectral data set for speed, We discuss here first the different file formats, and then explain the use of trafo.

Some theoretical background and technical details on the SPEX file format are given in Optimal definition of respons matrices and we refer the interested reader to that page for full details. Here we summarise some of the basics.

The basic philosophy behind the SPEX file format is:

Note

Keep things as simple as possible unless there is absolutely no alternative.

Therefore, SPEX uses one file containing all the information on the spectrum, including subtracted background, that must have the extension .spo and has FITS format with well-defined extensions and columns. Also, there is only a single file containing all the information on the response matrix, including effective area information, that must have the extension .res and also has FITS format with well-defined extensions and columns.

Note

Trafo always produces the .spo and the .res files at the same time, as both are linked together tightly.

Examples of the usage of trafo can be found in Analysis threads and How to convert spectra to SPEX format.

5.1.1. File formats spectra

There are a number of differences between OGIP and SPEX spectral file formats. See the OGIP Spectral file format specification.

Spectral format comparison

Property

OGIP

SPEX

Files

Separate files for source & optionally background and correction file

One file for all

Channel

2-byte or 4-byte column in spectral file

Not used (should be by definition the row number of the table, starting to count at 1 without gaps)

Bin boundaries

Real, but not in spectral file but in ebounds extension matrix

Double precision (for high-res spectra), and in spectral file

Data

Counts (2-byte or 4-byte integer) or count rate (real, in counts \mathrm{s}^{-1})

Only count rate (real, in counts \mathrm{s}^{-1})

Errors

Either explicitly given, or calculated using Poissonian statistics if poiserr keyword is given

Always explicitly given

Exposure (s)

One value for full spectrum

Can differ for each bin

Background

Unscaled value in separate file; needs backscal keyword or column in source & background file to determine scaling for source region

Background subtracted value in one column; subtracted background (and its statistical error) in other columns

Quality flag

2-byte integer with 4 different values (0:math:=good, 1=bad by s/w, 2=dubious by s/w, 5=set bad by user

“no-nonsense” logical, bin is either to be used (true) or not (false)

Grouping

2-byte integer, +1 for start channel, -1 for continuation bin, 0 if no grouping defined

Two logicals: true/false is bin is first/last of a group (i.e., for no grouping, both are true)

Area scaling

Areascal keyword or column

Not allowed (can be handled with exposure time or response matrix)

Background scaling

Backscal keyword or column

Worked out through explicit background subtraction in the spectrum

Systematic error

Keyword or column sys_err in both source & background files

Two columns: one for fraction of source, one for fraction of background

Table: Differences between OGIP and SPEX in spectral file structure

In Table Spectral format comparison we show the difference in structure between the OGIP spectral file format and the format used by SPEX. In addition to that, we make the following remarks.

For the energy grids, only energy (keV) units are allowed (not wavelength units), as there is a one-to-one relation between both. This number should be in double precision: for high-resolution spectra the difference between upper- and lower energy can be small, and close enough to the machine precision for single precision to cause some annoying numerical problems.

In the error column, it is essential to give real Poissonian errors (in case the spectrum is based on counts), and the use of e.g. Gehrels statistics must be avoided; in those cases, it is better to calculate the formal errors just from the square root of the number of raw counts. Chandra grating spectra are sometimes delivered with Gehrels errors, but this gives problems when the data need to be re-binned, and this is usually the case as the Chandra-data are delivered over-sampled. Also, never use 90 % errors, but always r.m.s. (“1\sigma”) errors.

For the same reasons, the use of systematic errors should be avoided in the spectral file, because after rebinning they would become smaller. It is better to use no systematic errors in the spectral file, but in case you really need them, within SPEX you can set them after you have done the proper binning.

Also the use of quality flags should be avoided. It is better to provide the user either with only “good” data, or to make the user aware of any systematic limitations of the dats. When needed (either for instrumental or astrophysical reasons), within SPEX it is possible to ignore data ranges.

The usefulness of the areascal keywords is not very clear; the XMM-Newton RGS software uses the arescal for some dead-time corrections, but SPEX solves this in a more elegant way by allowing the exposure time to be different for each data bin. Whenever trafo encounters the arescal, it uses it to adjust the exposure time per bin. If you give the “show data” command in , you see for each data set some statistics, including mean, minimum and maximum exposure time per bin.

Finally, OGIP uses three columns (the background count rate, and the backscal for the source and background region) to determine the background that needs to be subtracted. In SPEX, this reduces to two columns containing essentially the same information, namely the scaled background count rate and its statistical error. Actually, what is important in this is only the ratio of the backscal columns, not their individual values.

In summary, whenever possible we recommend to use only the first seven columns (bin boundaries, exposure time, source and background count rates with their errors), and leave the other columns empty / default (first/last channel flags, used, systematic errors).

5.1.2. File formats responses

The OGIP standard is described on this page.

Response format comparison

Property

OGIP

SPEX

Response:

Files

Separate files for response (.rmf) & ancillary response file (.arf)

One file for all

Rmf extension:

Components

1 component only

Matrix may have multiple components

Energy grid

One grid for the matrix, single precision

Each component can have its own grid, double precision

Response groups (\equiv contiguous row of non-zero matrix elements for the same energy)

Multiple groups per row; specify number of groups (2-byte), first channel & number of channels (2-byte or 4-byte) and the matrix elements for each group

In extension “spex_resp_group” specify bin lower and upper energy, first channel, last channel and number of channels for the group; in extension “spex_resp_resp give the matrix elements Optimalisation & No & Matrix may also contain derivatives of the matrix with respect to photon energy Ebounds extension: & & Channel energies & Single precision & Not here, but in spectral file and double precision Arf extension: & & Columns & Contains lower, upper energy and area (in cm^2) & N/A (included in matrix; but note units are SI, i.e. m^2)

Table: Differences between OGIP and SPEX in response matrix structure

In Table Response format comparison we show the difference in structure between the OGIP response file format and the format used by SPEX.

5.1.3. Multiple spectra

With trafo, you can combine different datasets into one combined spectrum and response file. There can be various reasons to do so:

  1. You may want to combine different, similar instruments from the same observation (e.g. RGS1 and RGS2) or different spectral orders (e.g. RGS1 -1 and RGS1 -2 spectral order), or even combine pn with RGS. However, in these cases you may prefer to have the spectra in separate files, as that allows you easier to use only part of the data.

  2. You may have time-resolved spectra of the same source taken with the same instrument

  3. You may have multiple spatial regions of the same source, observed with the same instruments.

For more info, we refer to Sectors and regions. Trafo allows you to achieve this.

5.1.4. How to use trafo

Trafo is an interactive program. It asks a few questions, which are usually self-explanatory. However, here we give a brief overview.

  1. The first question trafo asks if you want to transform data from OGIP format (option 1), the now abandoned SPEX version 1 binary format (option 2) or the new SPEX format (option 3).

  2. The next question is how many spectra you want to transform. Usually you will enter here the number 1, but if you want to create concantinated spectra (see previous section) you should enter here the number of spectra. In that case, the next questions will be repeated for each spectrum.

  3. Now you are asked to enter the maximum number of response groups per energy per spectrum. This is needed in order to allocate scratch memory; for almost any “sane” spectrum that we have seen sofar, a large number like 10000 is always sufficent here. Anyway, trafo will warn you and quit whenever you will reach this limit.

  4. Optional: for multiple spectra, you are asked how many sectors you want to create. See the description in Sect. [sec:sectorsandregions] for more info on sectors and regions.

  5. Optional: if you have more than 1 spectrum, trafo asks you to enter the sector and region number that you want to assign to the spectrum that you will enter next. If The sector or region number are out of the allowed range as specified before, trafo will keep repeating this question until you have entered valid numbers.

  6. Next a question is asked about partitioning of the matrix. You have here basically three options. Option 1: keep the structure of the matrix essentially as provided by the software package that created the OGIP files. The SPEX matrix will have 1 component, with no re-arrangements. Option 2: rearrange the matrix into contiguous groups. Your matrix may have been splitted into multiple regions, and for one photon energy you might have multiple regions (for instance, higher spectral orders for grating spectrometers without energy-sensitive detectors like the LETG/HRC-S or EUVE spectrometers); or a main diagonal and a fluorescence component, etc. Trafo will attempt to sort your response matrix accrding to these physically distinct components, by checking if for a given energy a response group has overlap in energy with an already existing group. Option 3: split into N roughly equal-sized components. This option is recommended for large matrices of high-resolution instruments such as RGS. It allows for the optimal usage of multiple processors during spectral fitting, provided your machine has of course more than one processor.

    Note that if you combine spectra that are already in SPEX version 2.0 or higher format, you do not have this option because we assume you have already optimised your matrix when creating the original version 2.0 files.

  7. Optional: if you have selected option 3 above, trafo ask you the number of components. This can be any number, but experience has shown that a power of 2, typically between 8 and 32 works best (even on dual core processors).

  8. Now you must enter the file name of the OGIP-type spectrum. Alternatively, if you combine SPEX version 2.0 files, you are prompted for the .spo file and you do not need to answer some of the questions below.

  9. If the OGIP spectral file does not provide you the name of a background spectrum, you are prompted if you want to subtract a background spectrum. Be aware that sometimes background subtraction has already been taken into account in OGIP-spectra. Check carefully. If you answer “yes” to this question, then trafo will ask you for the filename of the background file.

  10. In a few rare cases (e.g. Einstein SSS data), there is a second background file needed, the so-called “correction file”. If such a file is to be used, tarfo will read it but it must be indicated then as the corfil extension in the spectral file.

  11. Trafo now makes a few sanity checks. If there is some problem, trafo will report it and stop. It checks for the same number of data channels in source and background (or correction) file. Further, and this is important to know, data bins that are being qualified as “bad” in the background or correction files, but “good” in the source file, will end up as “bad” bins in the final, background subtracted spectrum.

  12. Trafo reports the total number of bad channels. You now can decide if you want to ignore the bad channels. Default is no (keep data; why would you otherwise have taken the burden to keep them in your datasets), but if you answer “yes”, the bad channels will be ignored by trafo (well, flaged as not to be used in the .spo file).

  13. Optional: if the OGIP-spectrum contains grouping information, trafo asks if that grouping should be used or ignored in the final spectrum.

  14. Optional: if there is no response matrix file specified in the spectral pha-file, trafo asks for the name of the matrix.

  15. Optional: while reading the response matrix, trafo makes some sanity checks. For instance, if the lower bin boundary of an energy bin is not smaller than the upper bin boundary, the user can correct this manually (some matrices are provided erroneously with zero width bins). But be sure that you understand here whjat you are doing!

  16. Optional: also, some instruments assign an energy 0 to the lower energy boundary of the first bin. SPEX does not like this (as it gives trouble if you convert to a wavelength scale, for instance), so you can change the lower boundary manually to a small, non-zero number here.

  17. Optional: in a few rare cases, matrices/data are poorly designed, such that the spectrum starts with a channel 0, but the matrix starts with channel 1. It is then not always clear which spectral element corresponds to which response element. Trafo tests for occurrences where the “flchan” keyword in the matrix equals 1, but the first channel in the data is 0. In this case it is possible to shift the response array by 1 channel, although this should be done as a last resort, and needs careefull checking if no mistakes are made! Trafo also tests for occurrences where the “flchan” keyword in the matrix does not equal 1 (usually 0), but the first channel in the data is 0. In this case it is advisable and possible to shift the response array by 1 channel, but again care should be taken!

  18. Optional: if there is no ancillary (effective area) file specified in the spectrum, trafo reports this and asks if the user wants to use such an arf-file nevertheless. Default is “no”, but if “yes” is entered, the name of the arf-file is prompted for.

  19. As a next step, model energy bins with zero effective area are deleted from the file. Such bins usually occur at the low and high-energy side of the matrix. Deleting them saves computing time. Further, any necessary rebinning (if indicated by the grouping falgs) is done.

  20. Trafo will tell you how many components it has found or created. It now asks you if you want to apply a shift to your spectra. Usually you should enter here 0. Useful cases to enter a non-zero shift s are situations where for instance your energy or wavelength scale is not yet sufficiently well calibrated. trafo then in that case puts the data of bin nr k+s into bin s.

  21. Trafo reports if the spectrum and matrix will be swapped (in case the original OGIP data were in wavelength order). Remember that SPEX always uses energy order.

  22. The pre-last question is the filename for the spectrum that has been created (without the .spo extension, that trafo will add automatically).

  23. Finally the filename for the response matrix that has been created is asked (without the .res extension, that trafo will add automatically).

5.1.5. Alternatives for trafo

Since we provide a package with Python tools for SPEX, there are a few scripts available that perform conversions from OGIP to SPEX format. ogip2spex is the program closest to the trafo program and can convert one spectrum from OGIP to SPEX format using only command line arguments.